中文摘要：

拉深过程中，压边力规律的实时预测是实现变压边力实时控制的关键。以盒形件为研究对象，引入人工神经网络，研究了变压边力规律实时预测模型的拓扑结构，获得了样本数据与提高预测模型收敛精度和泛化精度的方法。结果表明：对输入样本进行归一化处理，采用正则化训练方法，并适当地选取训练样本数和隐层节点数，可使预测模型的泛化精度控制在10％以内。

英文摘要：

The real time prediction of BHF control law in intelligent deep drawing is a key technology. A non--axisymmetrical workpiece such as rectangular box was studied and as a result the topological structure of BHF prediction model in real time was showed by introducing ANN. In this way preparion of sample, accuracy of model convergence and accuracy of model generalization were studied. It is indicated when input samples are normalized, training method is regularized and the right numbers of samples and hidden nodes are set, the accuracy of generalized results can be within 10%. This measure will meet to the engineering applications.